Conventional NAS Cluster Capacity Does Not Scale Efficiently.

Conventional NAS clusters do not use their capacity efficiently:

Cluster members that are islands:

When the cluster members are independent the capacity usage is uneven. As files cannot span hosts it becomes difficult to plan for growth so capacity growth is uneven with some hosts having greater capacity than others over time.

It is possible for a data set to run out of space in a host while the rest of the cluster has capacity to spare. It is even more inefficient to grow the capacity of all hosts evenly as this exacerbates the problem of the under-utilized hosts.

Clusters that do not scale out:

Some NAS clusters limit the number of hosts across which files can be striped. This limitation results in files not being to take full advantage of the cluster's capacity.

It is possible to have over one hundred hosts in such a cluster but any single file can only be striped across a dozen hosts. This creates a capacity imbalance as various subsets of the hosts of the cluster can grow at different rates.

Uneven capacity hinders resiliency:

For fault-tolerance purposes important files must be replicated to other hosts in the cluster. Because file size growth cannot always be accurately planned it is necessary to find replication hosts of equal capacity. The cost is multiplied while the rest of the cluster has capacity to spare. Selective host replication requires careful management and does not benefit the resiliency of the overall cluster. It becomes very hard to predict the impact of failures at any point in time. Data migration within the cluster becomes a complex and time consuming endeavor.

A Peer Fusion Cluster is a Tightly Integrated Group of Storage Peers.

A Peer Fusion NAS cluster is mathematically balanced to make optimal use of its capacity:

The cluster owns the data:

The data in a Peer Fusion cluster is striped across the cluster and onto every peer equally. This results in the even and optimal usage of the capacity of the cluster.

Growth is predictable based upon usage patterns and determining where to add capacity is simple: evenly across the cluster. It is also possible to grow the cluster's capacity by adding peers. This method of growth also improves the cluster's performance as it brings additional I/O, network, and CPU bandwidth.

Replication is not required:

There no longer is a need to replicate data and all data is protected against multiple concurrent peer failures. This results in more efficient usage of the cluster capacity as data does not consume multiple times its storage size. Not replicating files has the added advantage of improving cluster performance by eliminating the extra I/O required to maintain multiple copies of files.